Skip to content

Jan Nano

Jan Nano is a 4-billion parameter model designed for research and information retrieval. Instead of trying to know everything, it excels at finding anything through deep integration with Model Context Protocol (MCP) tools.

Two Variants

ModelContext WindowSizeUse Case
Jan Nano 32k32,768 tokens4-8GBQuick research, general queries
Jan Nano 128k131,072 tokens8-12GBDeep research, document analysis

What Makes Nano Different

  • Research-First Design: Trained to find relevant information, synthesize findings, and provide accurate citations.
  • MCP Integration: Works seamlessly with tools like web search, document analysis, and code repositories.
  • Extended Context: The 128k variant can process entire codebases, book-length documents, or 50+ research papers simultaneously.

Technical Details

  • Base: 4B parameter transformer
  • Training: Optimized for tool use and retrieval
  • Quantization: Q4, Q8, FP16 variants available

Philosophy

Most models try to be encyclopedias. Jan Nano is a research assistant. It doesn’t memorize the internet—it knows how to navigate it.